Deep learning architecture for automatic modulation classification in time-varying fading and impulsive noise channels

Abstract

The automatic modulation classification (AMC) allows identifying the kind of modulation of the received signal, being a key part of the development of cognitive radio devices that adapt the type of modulation to the characteristics of the communication environment. Several types of research on AMC have been done based on the analysis of the modulation signals and using its parameters for developing powerful feature descriptors to be used on this automatic classification. Recently, a new trend appears related to the use of architectures based on deep learning for this classification. Hence, in this work, we propose to use methods based on deep learning to classify the modulation type of a signal in an environment with doppler fading and impulsive noise. We studied and propose a model based on CNN that has shown to be comparable to the state-of-the-art methods.